1: \begin{abstract}

2: The effect of noise in the input data for learning potential energy

3: surfaces (PESs) based on neural networks for chemical applications is

4: assessed. Noise in energies and forces can result from aleatoric and

5: epistemic errors in the quantum chemical reference calculations.

6: Statistical (aleatoric) noise arises for example due to the need to

7: set convergence thresholds in the self consistent field (SCF)

8: iterations whereas systematic (epistemic) noise is due to, {\it inter

9:   alia}, particular choices of basis sets in the calculations. The two

10: molecules considered here as proxies are H$_{2}$CO and HONO which are

11: examples for single- and multi-reference problems, respectively, for

12: geometries around the minimum energy structure. For H$_2$CO it is

13: found that adding noise to energies with magnitudes representative of

14: single-point calculations does not deteriorate the quality of the

15: final PESs whereas increasing the noise level commensurate with

16: electronic structure calculations for more complicated,

17: e.g. metal-containing, systems is expected to have a more notable

18: effect. However, the effect of noise on the forces is more

19: noticeable. On the other hand, for HONO which requires a

20: multi-reference treatment, a clear correlation between model quality

21: and the degree of multi-reference character as measured by the $T_1$

22: amplitude is found. It is concluded that for chemically "simple" cases

23: the effect of aleatoric and epistemic noise is manageable without

24: evident deterioration of the trained model - although the quality of

25: the forces is important. However, considerably more care needs to be

26: exercised for situations in which multi-reference effects are present.

27: \end{abstract}

28: